Skip to content

WireFlow

Reproducible AI Workflows for Research & Development

Version 0.7.3 (pre-release)

Key Features

  • 🎯 Git-like Discovery: Run from anywhere in your project tree. WireFlow walks up to find .workflow/ automatically.

  • 📄 Native Documents: PDFs, Office files, images (including HEIC, TIFF, SVG) handled natively with automatic conversion.

  • 📎 Obsidian Embeds: ![[file]] syntax auto-resolves. Embedded images and PDFs become content blocks.

  • 🧠 Model Profiles: Switch between fast, balanced, and deep tiers. Enable extended thinking or effort levels for complex tasks.

  • 🔌 Multi-Provider: Use Anthropic Claude API or any OpenAI-compatible endpoint (LM Studio, ollama, vLLM).

  • 📦 Batch Processing: Process hundreds of documents at 50% cost savings with the Message Batches API.

  • 🔧 Config Cascade: Global → ancestors → project → workflow → CLI. Set once, override where needed.

  • 🏗️ Nested Projects: Inherit settings from parent projects. Perfect for monorepos.

  • 🔗 Workflow Chains: Build pipelines with --depends-on. Stale dependencies auto-execute before the target.

  • 📥 Input vs Context: Separate primary documents from supporting materials for cleaner prompts.

  • 💰 Prompt Caching: Smart ordering puts stable content first. Up to 90% savings on cached input tokens.

  • 📚 Citations: Enable source attribution with --enable-citations. Get references you can verify.

  • ⚡ Three Modes: Persistent workflows for iteration, quick task mode with hierarchical templates, or batch for bulk processing.

  • 📺 Streaming Output: Watch responses generate in real-time with incremental Markdown rendering in the terminal.

  • 🐚 Shell Integration: Bash completion, project-aware prompt (__wfw_ps1), and streamlined CLI.

  • 💾 Safe Outputs: Timestamped backups, hardlinked copies, atomic writes. Never lose work.

Quick Start

To install wireflow, clone the repo and run the installer:

# Clone repository
git clone https://github.com/jdmonaco/wireflow.git
cd wireflow

# Install (creates symlinks to ~/.local/bin and bash completions)
./wireflow.sh shell install

For Anthropic Claude API (default):

export ANTHROPIC_API_KEY="your-key"
export PATH="$HOME/.local/bin:$PATH"  # if not already in PATH

For local LLM servers (LM Studio, ollama, etc.):

# In ~/.config/wireflow/config
PROVIDER="openai"
OPENAI_BASE_URL="http://localhost:1234"
OPENAI_MODEL_BALANCED="your-model-name"

Try out initializing wireflow in any folder containing a project or files you want to process:

# Initialize project
cd ~/my-manuscript
wfw init .

# Create and run first workflow
wfw new 00-context-analysis
wfw run 00-context-analysis --stream

Your project files and folders are treated as read-only. All WireFlow files are maintained in a .workflow/ subfolder.

What You Can Do

Persistent Workflows: Create reusable workflows for iterative development:

wfw new analyze-data
wfw run analyze-data -cx "data/*.csv" --stream

One-Off Tasks: Execute lightweight tasks without workflow persistence:

wfw task -i "Extract key points" -cx notes.md
wfw task summarize -cx "*.md"

Workflow Chains: Build dependent pipelines with automatic context passing and stale dependency re-execution:

wfw run 02-analysis --depends-on 01-context --stream

Use Cases

  • Research manuscripts: Iterative analysis and writing assistance
  • Code analysis: Systematic code review and refactoring workflows
  • Documentation: Generate and update documentation from source code
  • Data processing: Multi-stage data analysis pipelines
  • Content generation: Structured content creation with context reuse

Documentation Structure

Requirements

  • Bash 4.0+
  • curl and jq for API interaction
  • Anthropic API key (get one here)

Optional: glow or gum for enhanced output display, yq for request inspection. See installation guide for details.

License

MIT License - see repository for details

Getting Help

wfw help              # Show all subcommands
wfw help <subcommand> # Show detailed help for a subcommand
wfw <subcommand> -h   # Quick help for a subcommand

Ready to get started? Head to the Installation Guide →